Has anyone implemented the new Morris Method by Campolongo and Braddock in 1999 (Reliability Engineering and Systems vol 64, pg 1-12)? This is a different approach to the conventional Morris Method and it allows first order interaction to be determined.
We have used it (or a variant) in a number of applications. The latest is:
https://www.researchgate.net/publication/257549855_Three_complementary_methods_for_sensitivity_analysis_of_a_water_quality_model?ev=prf_pub
Do you have specific question concerning the revised Morris Method?
Article Three complementary methods for sensitivity analysis of a wa...
Hi Barry,
Thank you for responding to my query. I can see that you implemented the original Morris method rather than the new Morris method developed by Campolongo and Braddock (1999) The use of graph theory in the sensitivity analysis of the model output: a second order screening method.
This method allows screening of second order interactions whereas the original Morris method, while it tells you higher order effects are important, does not tell you which of the higher order interactions are important.
My question regarding the new Morris method is on the sampling strategy, i.e. the way to construct the multiple trajectories that consists of joined single trajectory. Having followed the steps outlined by the paper, I am still not able to construct a multiple trajectory that can predict (k-1) distinct SEE per single orientation. I believe I may have misinterpreted the paper somewhere.
My apologies for not reading the paper you referred to before replying. I haven't used that version of the Morris method, and by the lack of other responses, it doesn't seem to have been used very much. Might be best if you contact one of the authors. Sorry I can't be more helpful, but your question has lead me to another variant of the Morris method, which I will look into.
I have managed to succesfully implemented the new Morris method. This is certainly an improvement on the original method given that it allows interaction effects to be assessed. The only downside is the longer runs needed (order ~rk^2). Depending on the type of problem and the number of factors, a full or fractional factorial 3 level design may be a better option in my opinion.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics